With its improved productivity and accuracy and more personalized experience, AI is revolutionizing medical imaging. According to Signify Research, the world market for AI in medical imaging — comprising software for automated detection, quantification, decision support, and diagnosis — will reach US$2 billion by 2023.
AI technologies utilized in medical imaging processing include deep learning, machine learning, AR, data mining, etc. These are capable of achieving a range of goals in disease screening, disease diagnosis, medical surgery and so forth.
AI Energizes Medical Image Processing: Medical image processing is essentially the application of computer vision technology. AI technologies such as machine learning and deep learning are used in the intelligent analysis of imaging data processed by computer vision technologies such as imaging registration and fusion; and can assist doctors in medical image labeling, disease diagnosis, and surgery.
Medical Images Integrate with Other Types of Data for Analysis: If medical imaging data is combined with physical signs of patients, medical history, genetic information, identity information and other non-image data in the training of AI algorithms, this can help the machine analyze the data in higher dimensions and extract the most essential characteristics. Exploring implied correlations behind the disease can assist doctors in diagnosing disease more accurately.
Deep Learning Exceeds Machine Learning in Disease Diagnosis: Conventional Computer Aided Diagnosis(CAD) methods are mainly based on machine learning or an expert system and have been criticized for running mechanically or being unhelpful. Deep learning algorithms however have proven more efficient, and can deal with medical data more comprehensively, extracting useful information and outputting key points of the disease, which liberates doctors from time-consuming clinical work.
AI in Medical Imaging Usage Scenarios
AI has advantages when dealing with medical imaging data, as CNN and RNN are adaptive for image processing spontaneously. In hospital and other medical institutions such as independent imaging centers and physical examination centers, medical imaging processing has become one of the most important practical applications of AI.
The table below excludes scenarios without medical imaging data, such as medical pure text processing. It also omits bioelectricity, where image processing technologies are dramatically different from general computer vision technologies.
Google — LYNA: Google’s LYmph Node Assistant (LYNA) deep learning program can be trained on pathology slides of patients with breast cancer to accurately detect the spread of breast cancer. The algorithm can distinguish slides with metastatic disease and pinpoint the site of cancers and other suspicious areas in each slide. The average time required for a pathologist to review a slide is only one minute with LYNA, compared to two minutes without the assistant.
ImmersiveTouch— ImmersiveView: This is a set of integrated VR real-time solutions for optimizing individualized surgical planning, patient engagement, and surgery teaching. The ImmersiveView suite transforms CT and MR images into intuitive, accurate, high-resolution VR models, allowing physicians to manipulate and explore the VR model of the patient to evaluate procedural options and prepare for surgery.
Intuitive Surgical — da Vinci System: The da Vinci System comprises a console and a patient-side cart containing a laparoscope, which is a thin tube with a micro camera and light at the terminal. Its design allows the surgeon to operate close to the console and move the camera, which sends images to a display to guide the surgeons.
Trends and Limitations for AI in Medical Imaging
AI technology has been welcomed by the medical imaging market as it can reduce inefficiencies and save doctors’ time, but there are also factors that limit the practical application of AI in medical imaging. For example, large amounts of medical imaging data are scattered across different hospitals, independent imaging centers and research institutions, which makes it difficult to assemble and utilize medical imaging data effectively.
Rapid improvements in AI continue to push medical imaging technology development and deployment. Apart from disease diagnosis, AI can also be applied in molecular/cellular level image processing and interventional imaging, assisting non-surgical diagnosis and treatment. It’s expected that regulators and industry associations will cooperate in the organization of medical imaging expert teams to establish algorithm model evaluation standards for the industry.
Author: Haitian Fu | Editor: Michael Sarazen